Efficient Adaptation of TTS Duration Model to New Speakers

Chilin Shih; Wentao Gu; Jan P.H. van Santen

Efficient Adaptation of TTS Duration Model to New Speakers

Chilin Shih, Wentao Gu, Jan P.H. van Santen

Research output: Contribution to conference › Paper › peer-review

Abstract

This paper discusses a methodology using a minimal set of sentences to adapt an existing TTS duration model to capture inter-speaker variations. The assumption is that the original duration database contains information of both language-specific and speaker-specific duration characteristics. In training a duration model for a new speaker, only the speaker-specific information needs to be modeled, therefore the size of the training data can be reduced drastically. Results from several experiments are compared and discussed.

Original language	English (US)
Pages	81-86
Number of pages	6
State	Published - 1998
Externally published	Yes
Event	3rd ESCA/COCOSDA Workshop on Speech Synthesis, SSW 1998 - Blue Mountains, Australia Duration: Nov 26 1998 → Nov 29 1998

Conference

Conference	3rd ESCA/COCOSDA Workshop on Speech Synthesis, SSW 1998
Country/Territory	Australia
City	Blue Mountains
Period	11/26/98 → 11/29/98

ASJC Scopus subject areas

Language and Linguistics
Cultural Studies

Library availability

Discover UIUC Full Text

Cite this

@conference{0e120e680a844d3fa5713c2ff109fa1d,

title = "Efficient Adaptation of TTS Duration Model to New Speakers",

abstract = "This paper discusses a methodology using a minimal set of sentences to adapt an existing TTS duration model to capture inter-speaker variations. The assumption is that the original duration database contains information of both language-specific and speaker-specific duration characteristics. In training a duration model for a new speaker, only the speaker-specific information needs to be modeled, therefore the size of the training data can be reduced drastically. Results from several experiments are compared and discussed.",

author = "Chilin Shih and Wentao Gu and {van Santen}, {Jan P.H.}",

note = "Publisher Copyright: {\textcopyright} 1998 3rd ESCA/COCOSDA Workshop on Speech Synthesis, SSW 1998. All rights reserved.; 3rd ESCA/COCOSDA Workshop on Speech Synthesis, SSW 1998 ; Conference date: 26-11-1998 Through 29-11-1998",

year = "1998",

language = "English (US)",

pages = "81--86",

}

TY - CONF

T1 - Efficient Adaptation of TTS Duration Model to New Speakers

AU - Shih, Chilin

AU - Gu, Wentao

AU - van Santen, Jan P.H.

PY - 1998

Y1 - 1998

N2 - This paper discusses a methodology using a minimal set of sentences to adapt an existing TTS duration model to capture inter-speaker variations. The assumption is that the original duration database contains information of both language-specific and speaker-specific duration characteristics. In training a duration model for a new speaker, only the speaker-specific information needs to be modeled, therefore the size of the training data can be reduced drastically. Results from several experiments are compared and discussed.

AB - This paper discusses a methodology using a minimal set of sentences to adapt an existing TTS duration model to capture inter-speaker variations. The assumption is that the original duration database contains information of both language-specific and speaker-specific duration characteristics. In training a duration model for a new speaker, only the speaker-specific information needs to be modeled, therefore the size of the training data can be reduced drastically. Results from several experiments are compared and discussed.

UR - http://www.scopus.com/inward/record.url?scp=85133518145&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85133518145&partnerID=8YFLogxK

M3 - Paper

AN - SCOPUS:85133518145

SP - 81

EP - 86

T2 - 3rd ESCA/COCOSDA Workshop on Speech Synthesis, SSW 1998

Y2 - 26 November 1998 through 29 November 1998

ER -

Efficient Adaptation of TTS Duration Model to New Speakers

Abstract

Conference

ASJC Scopus subject areas

Library availability

Related links

Fingerprint

Cite this